A central problem in automated reasoning is how to cope with uncertainty. The paper describes a methodology for symbolic uncertain inference, making a distinction between uncertain data and uncertain inferences. Uncertainty is represented by higher-order quantifiers called “modulations,” as opposed to the modal quantifiers of first-order modal logic. The approach works in the framework of the most common deductive inference schema--the resolution principle. Modulations are composed during the inference on the clauses and kept explicitly.

The paper contains some important ideas concerning inexact reasoning. Emphasis is put on fundamental principles. The text is intuitive, and flavored by numerous examples which make it readable. Details of the way the modulations compositions must be made are not given. Considerable work is left to those who intend to introduce inexact reasoning in expert systems.